def get_fc7(graph_filename, load_filename):
graph = load_graph(graph_filename)
#for op in graph.get_operations():
# print(op.name, op.values())
input = graph.get_tensor_by_name('prefix/input:0')
prob = graph.get_tensor_by_name('prefix/test/prob:0')
fc7 = graph.get_tensor_by_name('prefix/fc7/fc7:0')
keep_prob = graph.get_tensor_by_name('prefix/Placeholder_1:0')
paths, labels = utils.load_paths(load_filename, './')
features = []
with tf.Session(graph=graph) as sess:
# with tf.device('gpu:/0'):
for i in range(len(paths)):
path.DataDir.percent_bar(i + 1, len(paths))
#print(paths[i])
images = utils.load_inputs(paths[i])
#print(images)
images = np.asarray(images, dtype=np.float32)
out = sess.run(fc7, feed_dict={input: images, keep_prob: 1.})
features.append(out)
#print('Gain %d utterance feature' % i)
return features, labels
def main(date, hour):
if date == 'n':
date = None
if hour == 'n':
hour = None
abspath = os.path.abspath(__file__)
dname = os.path.dirname(abspath)
os.chdir(dname)
dir = os.getcwd()
paths = ut.load_paths(dir)
paths['output'] = os.path.abspath(paths['output'])
paths['data_store'] = os.path.abspath(paths['data_store'])
pull_gefs_files(date=date, hour=hour)
fmean, fsprd = run_fcsts(paths=paths)
date = pd.to_datetime(fmean['valid_time'][0].values)
logging.info('mcli started')
mc_mean, mc_std = run_mcli()
mc_std = mc_std.dropna(dim='lat')
mc_mean = mc_mean.dropna(dim='lat')
fmean, fsprd = align_fmean_fsprd(fmean, fsprd, mc_mean)
fmean.to_netcdf(f'{paths["output"]}/slp_mean_{date.year}{date.month:02}{date.day:02}_{date.hour:02}z.nc')
fsprd.to_netcdf(f'{paths["output"]}/slp_sprd_{date.year}{date.month:02}{date.day:02}_{date.hour:02}z.nc')
logging.info('fmean and spread saved, mcli finished')
logging.info('percentile started')
percentile = combine_fcast_and_mcli(fmean, mc_mean)
gc.collect()
logging.info('percentile complete')
subset_sprd = transforms.subset_sprd(percentile, mc_std)
logging.info('spread subset complete')
hsa_final, ssa_perc = transforms.hsa(fsprd, subset_sprd)
ssa_perc.to_netcdf(f'{paths["output"]}/ssa_perc_{date.year}{date.month:02}{date.day:02}_{date.hour:02}z.nc')
hsa_final.to_netcdf(f'{paths["output"]}/hsa_{date.year}{date.month:02}{date.day:02}_{date.hour:02}z.nc')
logging.info('hsa and ssa percentile file created')
def grid_search(model_generator, param_grid, feature_set):
print('\nGrid Search')
# load data
print('loading', feature_set)
paths = u.load_paths('PATHS.yaml') # get paths from file
train_x, val_x, test_x = u.load_data(paths['extracted_data'] +
'features_%s.p' % feature_set)
train_y, val_y, test_y = u.load_data(paths['extracted_data'] +
'labels_%s.p' % feature_set)
test_x = np.vstack(
(val_x,
test_x)) # dont use validation set for grid search, add to test data
test_y = np.vstack((val_y, test_y))
# train
# grid search train with 3 fold validation
classifier = KerasClassifier(model_generator,
n_dim=train_x.shape[1:],
n_labels=test_y.shape[1])
validator = GridSearchCV(classifier,
param_grid=param_grid,
scoring='neg_log_loss',
verbose=3,
n_jobs=1)
validator.fit(train_x, train_y)
# results
search_results(validator, test_x, test_y)
def random_search(model_generator, param_dist, n_iter, feature_set):
print('\nRandomized {[Search witn %d iterations\n' % n_iter)
# load data
print('loading', feature_set)
paths = u.load_paths('PATHS.yaml') # get paths from file
train_x, val_x, test_x = u.load_data(paths['extracted_data'] +
'features_%s.p' % feature_set)
train_y, val_y, test_y = u.load_data(paths['extracted_data'] +
'labels_%s.p' % feature_set)
test_x = np.vstack(
(val_x,
test_x)) # dont use validation set for grid search, add to test data
test_y = np.vstack((val_y, test_y))
# train
# grid search train with 3 fold validation
classifier = KerasClassifier(model_generator,
n_dim=train_x.shape[1:],
n_labels=test_y.shape[1])
validator = RandomizedSearchCV(classifier,
param_distributions=param_dist,
n_iter=n_iter,
verbose=3)
validator.fit(train_x, train_y)
# results
search_results(validator, test_x, test_y)
def multiplier(IN_PATH, OUT_PATH, single=False, subfolders=True):
# Tableau de chemins des fichiers images a traiter
if single:
images_path = [IN_PATH]
else:
images_path = load_paths(IN_PATH)
# Creer le dossier de sortie s'il n'existe pas
makedirs(OUT_PATH, exist_ok=True)
print('> Multiplication')
print('- Dossier entree : "%s"' % IN_PATH)
print('- Dossier sortie : "%s"' % OUT_PATH)
transform(images_path, OUT_PATH, copy, not single or subfolders)
transform(images_path, OUT_PATH, flip, not single or subfolders)
transform(images_path, OUT_PATH, rotation_90, not single or subfolders)
transform(images_path, OUT_PATH, rotation_180, not single or subfolders)
# perte couleurs
transform(images_path, OUT_PATH, grayscale, not single or subfolders)
transform(images_path, OUT_PATH, invert, not single or subfolders)
# perte pixels
transform(images_path, OUT_PATH, crop, not single or subfolders)
transform(images_path, OUT_PATH, blue, not single or subfolders)
#multiplier('../donnees-projet/Data', './sortie/multiplication')
def __init__(self,
stat: str,
variable: str,
paths: dict = None,
group: bool = False):
self.variable = self.convert_variable(variable)
if stat == 'mean':
stat = 'geavg'
elif stat == 'sprd' or stat == 'std':
stat = 'gespr'
else:
raise ValueError('Stat must be mean or sprd/std')
self.stat = stat
if paths == None:
self.paths = ut.load_paths()
else:
self.paths = paths
if group == True:
self.load_all()
def get_data(feature_set):
# load dataset
paths = u.load_paths('PATHS.yaml') # get paths from file
train_x, val_x, test_x = u.load_data(paths['extracted_data'] +
'features_%s.p' % feature_set)
train_y, val_y, test_y = u.load_data(paths['extracted_data'] +
'labels_%s.p' % feature_set)
test_x = np.vstack(
(val_x,
test_x)) # dont use validation set for grid search, add to test data
test_y = np.vstack((val_y, test_y))
train_y = u.inv_one_hot_encode(train_y) # remove one hot encoding
test_y = u.inv_one_hot_encode(test_y)
if feature_set == u.FEATURE_SET_SPECS_NORM:
# flatten 128 x 128 image
length = train_x.shape[1] * train_x.shape[2]
train_x = train_x.reshape(train_x.shape[0], length)
test_x = test_x.reshape(test_x.shape[0], length)
data = {
'train': {
'X': train_x,
'y': train_y
},
'test': {
'X': test_x,
'y': test_y
},
'n_classes': len(np.unique(train_y))
}
print("dataset has %i training samples and %i test samples." %
(len(data['train']['X']), len(data['test']['X'])))
return data
def train_model(model_generator, feature_set):
# TODO get data function that returns dict (options for one hot or not, val or not)
# load dataset
paths = u.load_paths('PATHS.yaml') # get paths from file
train_x, val_x, test_x = u.load_data(paths['extracted_data'] +
'features_%s.p' % feature_set)
train_y, val_y, test_y = u.load_data(paths['extracted_data'] +
'labels_%s.p' % feature_set)
model_name, model, training = model_generator(n_dim=train_x.shape[1:],
n_labels=test_y.shape[1])
run_id = '%s_%s' % (model_name, datetime.datetime.now().isoformat())
print('\nTrain and Evaluate: %s' % model_name)
# callbacks
earlystop = EarlyStopping(monitor='val_loss',
patience=training.early_stop_patience,
verbose=1,
mode='auto')
log_dir = os.path.join(paths['tensorboard_logs'], run_id)
tensorboard = TensorBoard(log_dir=log_dir,
histogram_freq=3,
write_graph=True)
t0 = time.time()
history = model.fit(train_x,
train_y,
validation_data=(val_x, val_y),
callbacks=[earlystop, tensorboard],
nb_epoch=training.n_epoch,
batch_size=training.batch_size)
training_time = time.time() - t0
# test
y_prob = model.predict_proba(test_x, verbose=0)
y_pred = np_utils.probas_to_classes(y_prob)
y_true = np.argmax(test_y, 1)
# evaluate the model's accuracy
t0 = time.time()
score, accuracy = model.evaluate(test_x,
test_y,
batch_size=training.batch_size)
testing_time = time.time() - t0
cm = confusion_matrix(y_true, y_pred, labels=None)
# p, r, f, s = precision_recall_fscore_support(y_true, y_pred, average='micro')
# roc = roc_auc_score(test_y, y_prob)
# print("F-Score:", round(f, 2)) # similar value to the accuracy score, but useful for cross-checking
# print("ROC:", round(roc, 3))
# print results
print("\nclassifier: %s" % model_name)
print("training time: %0.4fs" % training_time)
print("testing time: %0.4fs" % testing_time)
print("accuracy: %0.4f" % accuracy)
print("confusion matrix:\n%s" % cm)
# print model.summary()
# plot and save results
fname = paths['model_save'] + model_name + '_accuracy_%0.2f' % accuracy
u.plot_keras_loss(fname, history) # saves plot png
model.save(fname + '.h5')
cm_path = './confusion_plots/%s' % model_name
cm_title = '%s (Accuracy: %0.2f)' % (model_name, accuracy)
u.plot_confusion_matrix(cm, cm_path, title=cm_title)
# create database
known_names, known_faces_encoding = utils.create_database(
'../Images/database/group1')
# load images to test
unknown_names, unknown_faces_encoding = utils.create_database(
'../Images/test_face/group1')
known_names.append('unknown')
index = [
utils.recognize_face(unknown_face_encoding, known_faces_encoding)
for unknown_face_encoding in unknown_faces_encoding
]
names = [known_names[i] for i in index]
utils.compare_result(unknown_names, names)
# results :
# test only group 1 : Result : 1 error out of 18. Accuracy: 0.94
# test only group 2 : Result : 0 error out of 8. Accuracy: 1
# test group 3 (all photos) : Result : 1 error out of 26. Accuracy: 0.96
filepath = '../Images/test_mouth/'
predictor_path = './shape_predictor_68_face_landmarks.dat'
paths = utils.load_paths(filepath)
image_paths = [os.path.join(filepath, f) for f in paths]
for image_path in image_paths:
mouth_detection.mouth_location_image(image_path, predictor_path)
save_features_Up(features_Vp, weights, train_utterance_file[i])
print('save speaker %s train features utterance' %
DataDir.val_speaker[i])
test_features = np.load(test_features_file[i])
features_Vp = get_features_Vp(test_features)
save_features_Up(features_Vp, weights, test_utterance_file[i])
print('save speaker %s test features utterance' %
DataDir.val_speaker[i])
# solve weights
elif sys.argv[1] == '-w':
print('solve_w')
train_features = np.load(train_features_file)
features_Vp = get_features_Vp(train_features)
paths, labels = utils.load_paths(train_filename, './')
labels = np.asarray(labels)
print(features_Vp.shape, labels.shape)
weights, features_Up = solve_weights(features_Vp, labels)
weights = np.array([1])
save_features_Up(features_Up, weights, train_utterance_file)
test_features = np.load(test_features_file)
features_Vp = get_features_Vp(test_features)
paths, labels = utils.load_paths(test_filename, './')
labels = np.asarray(labels)
print(features_Vp.shape, labels.shape)
weights, features_Up = solve_weights(features_Vp, labels)
weights = np.array([1])
save_features_Up(features_Up, weights, test_utterance_file)
# -*- coding: utf-8 -*-
import sys
sys.path.insert(1, "../traitement-images")
from os import path, makedirs
from PIL import Image as pImage
from PIL import ImageOps
from utils import load_paths, add_suffix
import itertools
from filtres import *
IN_PATH = './200x200'
#IN_PATH = './test'
OUT_PATH = './dataset_exp'
images_origin = load_paths(IN_PATH)
def transform(outpath, paths, fnc):
result = []
for image_path in paths:
image = pImage.open(image_path)
print('"%s" => %s' % (image_path, fnc.__name__))
transformed = fnc(image)
in_path = path.dirname(path.dirname(image_path))
transformed_path = add_suffix(in_path, outpath, image_path,
'_' + fnc.__name__)
import utils as u
paths = u.load_paths('PATHS.yaml') # get paths from file
AUDIO_DIR = paths['audio_data']
SAVE_DIR = paths['extracted_data']
TRAIN = [
'fold1', 'fold2', 'fold3', 'fold4', 'fold5', 'fold6', 'fold7', 'fold8'
]
VAL = ['fold9']
TEST = ['fold10']
SAMPLE = ['fold_sample']
def dump_features(name, extractor):
print('Extracting %s' % name)
fnames = ['features_' + name, 'labels_' + name]
train_features, train_labels = extractor(AUDIO_DIR, TRAIN)
val_features, val_labels = extractor(AUDIO_DIR, VAL)
test_features, test_labels = extractor(AUDIO_DIR, TEST)
data = [(train_features, val_features, test_features),
(train_labels, val_labels, test_labels)]
u.dump_data(SAVE_DIR, data, fnames)
def dump_sample(name, extractor):
print('Extracting %s' % name)
fnames = ['features_' + name, 'labels_' + name]
samp_features, samp_labels = extractor(AUDIO_DIR, SAMPLE)
data = [(samp_features), (samp_labels)]
u.dump_data(SAVE_DIR, data, fnames)